Statistical Learning across Representational Levels: Neural and Behavioral Expression, Modality, Individual Variability and Lifespan

Statistical learning (SL) refers to the ability to extract regularities from structured sensory input and has been proposed as a fundamental mechanism supporting language acquisition, reading, and adaptive behavior. Despite extensive research, important questions remain regarding the nature of SL, its organization across modalities, its development across the lifespan, and the relationship between neural and behavioral expressions of learning. The present thesis addressed these questions through a series of behavioral and imaging studies examining SL in children, young adults, and older adults. Across the empirical chapters, SL was investigated using both neural frequency-tagging paradigms and behavioral measures, with a particular focus on modality-specificity, cross-modal transfer, individual variability, and lifespan changes. Results revealed robust neural sensitivity to structured regularities in both young and older adults, even when explicit behavioral evidence of learning was weak or absent. Cross-modal studies further demonstrated that learning and transfer varied across auditory and visual domains, suggesting that SL is shaped by both shared computational principles and modality-specific representational constraints. Developmental and aging findings indicated that some forms of sensitivity to regularities remain relatively preserved across the lifespan, although the behavioral expression of learning and the integration of information across modalities may differ as a function of age. Across studies, substantial inter-individual variability emerged, and correlations between neural and behavioral measures were often limited. These findings challenge strongly unitary conceptions of SL and instead support the view that learning is best understood as a multicomponential phenomenon. Building on these results, the thesis proposes a multi-level framework in which different paradigms capture distinct expressions of learning that vary in representational format, processing demands, and accessibility to conscious report. Within this framework, neural entrainment primarily reflects early online sensitivity to statistical structure, whereas behavioral measures additionally recruit processes related to prediction, memory retrieval, decision-making, and metacognitive reflection. Overall, the present work argues that SL is neither fully domain-general nor entirely modality-specific, but rather emerges from dynamic interactions among perceptual systems, prior knowledge, developmental factors, and individual cognitive differences. By integrating neural and behavioral methodologies across modalities and age groups, the thesis contributes to a more comprehensive understanding of how humans acquire and generalize statistical structure in complex sensory environments.